Simular $21.5M Series A: Desktop AI Agents Solve Hallucination Through Deterministic Workflows

Simular raised $21.5 million in Series A funding from Felicis to scale desktop AI agents that control entire Mac and Windows computers directly—solving the hallucination problem that has kept enterprise AI automation confined to simple browser tasks.

The fundamental constraint limiting AI agent enterprise adoption isn’t capability but reliability. Current large language models hallucinate unpredictably, and when agents execute thousands of desktop steps, even small errors cascade into complete workflow failures. Simular’s breakthrough addresses this through “neuro symbolic computer use agents” that let AI explore freely, then lock successful workflows into deterministic code.

The Desktop Control Infrastructure Gap

Browser-based automation dominated the first wave of AI agents, but enterprise workflows require control across multiple desktop applications—spreadsheets, databases, proprietary software, and operating system functions. Traditional solutions force deterministic behavior that kills creativity, or accept unreliable performance that fails enterprise standards.

Simular’s platform provides direct mouse and keyboard control across any desktop application. “We can literally move the mouse on the screen and do the click,” CEO Ang Li explained to TechCrunch. This capability extends beyond web automation into full desktop orchestration—exactly what enterprises need for comprehensive workflow automation.

The Microsoft validation came immediately: Simular was selected as one of five companies for the Windows 365 for Agents program, alongside Manus AI, Fellou, Genspark, and TinyFish. This partnership signals Microsoft’s serious commitment to agentic computing infrastructure.

The Deterministic Workflow Solution

Simular’s technical breakthrough lies in hybrid execution: AI agents explore and iterate freely with human feedback, but once successful workflows emerge, the platform converts them into deterministic code. This eliminates the reliability vs. creativity tradeoff that has constrained enterprise AI adoption.

Li and co-founder Jiachen Yang, both reinforcement learning specialists from Google DeepMind’s product teams (including Waymo development), engineered this “neuro symbolic” approach specifically for production environments. “Let agents keep exploring the successful trajectory. Once you found a successful trajectory, that becomes deterministic code,” Li explained.

The deterministic conversion puts final workflow code directly in users’ hands, enabling inspection, auditing, and enterprise trust. This addresses the governance requirements that have blocked AI agent deployment in regulated industries and risk-averse enterprises.

Enterprise Adoption Evidence

Early beta customers validate Simular’s enterprise applicability across industries requiring complex multi-system workflows:

  • Automotive: Car dealerships automating VIN number lookups across multiple databases and systems
  • Real Estate: HOAs extracting and processing contract details from PDF documents
  • Content Operations: Automated content creation pipelines spanning multiple applications

The open source project has spawned additional automations ranging from complex sales workflows to multi-step data processing pipelines. This organic adoption demonstrates market demand for reliable desktop automation infrastructure.

The investment backing reinforces enterprise focus: NVentures (NVIDIA’s venture arm) doubled down from the $5 million seed round, joined by South Park Commons, Basis Set Ventures, Flying Fish Partners, Samsung NEXT, and angel investor Lenny Rachitsky.

Market Infrastructure Shift

Simular’s approach represents evolution from browser automation to full computing environment control. While competitors focus on web-based tasks, the real enterprise value lies in automating repetitive workflows across all desktop applications—the majority of knowledge worker productivity.

This desktop control infrastructure could become foundational as Microsoft pushes agent integration, Google develops rumored desktop AI features, and Apple advances AI capabilities. The deterministic workflow approach positions Simular for enterprise segments where reliability requirements exceed creative flexibility needs.

The $21.5 million Series A (total $27 million raised) validates that desktop-level AI control represents the next infrastructure layer for workplace automation. Enterprise customers are clearly willing to pay premium for reliable automation that works across their entire software stack, not just web applications.

Looking Forward: Post-Browser Computing

The timing aligns with broader industry shifts toward agentic computing. AWS introduced new agent infrastructure, IBM partnered with AWS on enterprise agentic AI, and Microsoft’s Windows 365 for Agents program signals platform-level commitment to AI automation.

Simular’s deterministic approach could accelerate enterprise adoption by solving the fundamental trust barrier. When IT departments can inspect, audit, and trust AI-generated automation code, deployment resistance decreases significantly. This infrastructure foundation enables scaling beyond pilot programs into business-critical operations.

The desktop automation market will likely consolidate around reliability rather than capability demonstrations. Simular’s hybrid model—creative exploration plus deterministic execution—provides the governance and auditability that enterprises require for large-scale AI agent deployment.

Desktop AI agent infrastructure represents the bridge between current browser automation limitations and comprehensive enterprise workflow orchestration. Simular’s deterministic approach addresses the core reliability barriers that have constrained AI agent adoption in business-critical environments. This type of infrastructure development enables the transition from experimental automation to production-scale enterprise deployment—exactly the foundation needed for workplace AI transformation.

For organizations building agentic workflows across multiple applications and environments, platforms like Overclock provide the orchestration layer to coordinate these emerging desktop automation capabilities within broader business process architecture.